An evolutionary feature selection system and method that determines a feature space for a dataset. A system is disclosed that includes: a system for generating a plurality of chromosomes; an agglomerative K-means clustering system for clustering data into clusters, wherein each of the cluster spaces is associated with a different one of the chromosomes; a linear discriminant analysis system for scoring each of the cluster spaces; and an evolutionary mating system that genetically mutates and mates at least two of the chromosomes associated with the highest scoring cluster spaces, and generates a final chromosome. The final chromosome can thereafter be used to define a feature space in a matching system that attempts to match inputted biometric data with entries in a biometric dataset.
Legal claims defining the scope of protection, as filed with the USPTO.
1. An evolutionary feature selection system for determining a meta-feature space, comprising: a computing device, including: a system for generating a plurality of chromosomes; an agglomerative K-means clustering system for generating a cluster space for each of the plurality of chromosomes, wherein the generating of the cluster space includes evaluating data elements for cluster membership based on a minimum Euclidean distance between a feature vector of a data element and a multi-dimensional centroid of each cluster, wherein each cluster space is associated with a different one of the chromosomes and includes a set of clusters; a linear discriminant analysis system for scoring each of the cluster spaces, wherein the linear discriminant analysis system determines a score based on a ratio maximization of between class scatter and within class scatter; and an evolutionary mating system that performs the following: genetically mutates and mates chromosomes in a pairwise fashion based on the scoring of associated cluster spaces, wherein the genetically mutating and mating of the chromosomes includes continuously mating a top two scoring chromosomes in the chromosomes as a pair; and generates a final chromosome that defines a meta-feature space after completion of the continuously mating.
2. The evolutionary feature selection system of claim 1 , wherein each chromosome comprises a binary string having a plurality of bits, wherein each bit represents a presence or absence of a biometric trait.
3. The evolutionary feature selection system of claim 2 , wherein each biometric trait represents a facial feature.
4. The evolutionary feature selection system of claim 1 , wherein the data comprises a dataset of facial features.
5. The evolutionary feature selection system of claim 1 , wherein the plurality of chromosomes are generated randomly.
6. The system of claim 1 , wherein the evolutionary mating system continuously mates the top two scoring chromosomes for a predefined number of generations.
7. A non-transitory computer readable medium having a program product stored thereon for defining a meta-feature space from a dataset, comprising program code for: generating a plurality of chromosomes; clustering data from the dataset into cluster spaces using agglomerative K-means clustering, wherein the clustering includes evaluating data elements for cluster membership based on a minimum Euclidean distance between a feature vector of a data element and a multi-dimensional centroid of each cluster, wherein each cluster space is associated with a different one of the chromosomes and includes a set of clusters; scoring each of the cluster spaces using linear discriminant analysis, wherein the linear discriminant analysis determines a score based on a ratio maximization of between class scatter and within class scatter; and genetically mutating and mating chromosomes in a pairwise fashion based on scoring of associated cluster spaces, wherein the genetically mutating and mating of the chromosomes includes continuously mating a top two scoring chromosomes in the chromosomes as a pair; and outputting a final chromosome that defines a meta-feature space after completion of the continuously mating.
8. The computer readable medium of claim 7 , wherein each chromosome comprises a binary string having a plurality of bits, and wherein each bit represents a presence or absence of a biometric trait.
9. The computer readable medium of claim 8 , wherein each biometric trait represents a facial feature.
10. The computer readable medium of claim 7 , wherein the data comprises a dataset of facial features.
11. The computer readable medium of claim 7 , wherein the plurality of chromosomes are generated randomly.
12. The program product of claim 7 , wherein the continuously mating of the top two scoring chromosomes is performed for a predefined number of generations.
13. A method of determining a meta-feature space from a dataset, comprising: generating a plurality of chromosomes; clustering data from the dataset into cluster spaces using agglomerative K-means clustering, wherein the clustering includes evaluating data elements for cluster membership based on a minimum Euclidean distance between a feature vector of a data element and a multi-dimensional centroid of each cluster, wherein each cluster space is associated with a different one of the chromosomes and includes a set of clusters; scoring each of the cluster spaces using linear discriminant analysis, wherein the linear discriminant analysis determines a score based on a ratio maximization of between class scatter and within class scatter; genetically mutating and mating the chromosomes in a pairwise fashion based on scoring of associated cluster spaces, wherein the genetically mutating and mating of the chromosomes includes continuously mating a top two scoring chromosomes in the chromosomes as a pair; and outputting a final chromosome that represents a meta-feature space after completion of the continuously mating.
14. The method of claim 13 , wherein each chromosome comprises a binary string having a plurality of bits, and wherein each bit represents a presence or absence of a biometric trait.
15. The method of claim 14 , wherein each biometric trait represents a facial feature.
16. The method of claim 13 , wherein the dataset comprises a dataset of facial features.
17. The method of claim 13 , wherein the plurality of chromosomes are generated randomly.
18. The method of claim 13 , further comprising: submitting the final chromosome to a matching system; using the chromosome for matching inputted biometric data with a dataset of biometric features.
19. The method of claim 13 , wherein the continuously mating of the top two scoring chromosomes is performed for a predefined number of generations.
20. A method for deploying a system determining a feature space from a dataset, comprising: providing a computer infrastructure being operable to: generate a plurality of chromosomes; cluster data into cluster spaces using agglomerative K-means clustering, wherein the clustering includes evaluating data elements for cluster membership based on a minimum Euclidean distance between a feature vector of a data element and a multi-dimensional centroid of each cluster, wherein each cluster space is associated with a different one of the chromosomes and includes a set of clusters; score each of the cluster spaces using linear discriminant analysis, wherein the linear discriminant analysis determines a score based on a ratio maximization of between class scatter and within class scatter; and genetically mutate and mate at least two of the chromosomes associated with highest scoring cluster spaces, wherein the genetically mutating and mating of the at least two of the chromosomes includes continuously mating a top two scoring chromosomes in the chromosomes as a pair; and output a final chromosome after completion of the continuously mating.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
June 11, 2008
May 29, 2012
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.